Models of what the brain does with the world.
Mind Matter Labs is building a foundation model of the human brain: given a piece of media, it predicts the response that media would produce across human cortex.
Who we are
Computer science student at the University of Pennsylvania, with an interest in the human brain for as long as memory serves.
AI Engineering student at the University of Pennsylvania, with machine-learning research experience at Penn's GRASP Lab.
What we're building
A brain-encoding model takes a stimulus and predicts the neural response to it. Ours are fitted on fMRI datasets, where people looked at images, watched video and listened inside a scanner while their cortex was recorded. A foundation model brings those together: one model of how the brain takes in what people see, hear and read.
How we evaluate it
The hard part of this field is not fitting a model, it is knowing whether it generalises. Ours is tested on people it was never fitted on and stimuli it never saw, not on held-out clips from subjects it already knows, which is a much easier test and a much less useful one.
We are equally interested in where it fails. It tells you how the brain responds, not whether anyone will like what they see, click it, or buy anything. We would rather say that plainly than have you find out.
Where we are
Early, and backed by a16z speedrun. We are working with a small group of early partners. If you have a use for this we did not anticipate, we would genuinely like to hear about it.